Proof Systems

Lecture 3 Logic and Programming Languages Proof system

• A proof system is a formalized system for proving things. • Most systems have several components 1. A set of Axioms. Things that are known to be true without any work 2. A set if inference rules for deriving larger true statements from smaller true statements 3. A set of assumptions from which to work 1. In a mechanized logic, a proof is a data structure that can be checked by a machine Consistency, completeness, normal forms • Consistency – A system is consistent if falsehood is not provable (from the empty set of assumptions) – A system is complete if every theorem is provable from the inference rules of the logic – A Normal Form exists of there exists a unique smallest proof for every theorem, and other proofs of the same theorem “reduce” to this proof. Natural Deduction

• A style of proof with several elements that have become widely used 1. Introduction rules 2. Elimination rules 3. Hypothetical judgements 1. Reasoning from assumptions

1. Proofs are represented by a tree of “true statements” rooted at the bottom. Proof trees • A proof tree has several parts 1. A statement of what is proven (the root). Drawn below the line 2. A set of sub trees that represent proofs of the required components. Drawn above the line 3. A name for the inference rule used. Draw to the left of the line. 4. A set of premises. Drawn in brackets

[p0 /\ p1] ------/\e2 p1 [p2] ------/\i p1 /\ p2 ------~~i ~~(p1 /\ p2) Introduction rules

• For each connective of the logic, there is an introduction rule, where the root (below the line) has that connective has its outermost form. Elimination rules

• For each connective there is a rule that tells how to “consume” a formula with that connective to prove something else. Here the formula with that connective is above the line. Hypothetical Judgements

• Somethings can be proven using a sort of conditional reasoning. • We need a way to “temporarily” assume a new condition, and then cut of this assumption when we are done. – Assume some formula are true – Infer other things follow from these assumptions • These are consequences of the assumptions Natural deduction by the rules

• We will look at each connective, and then study both the introduction and elimination rules for it. And Or Not Implies Semantics

• The statement below the line is a consequence of the premises, and if it is in a box, the assumption of the box. • Natural deduction works by maintaining this invariant • Every step keeps the invariant true Natural Deduction as a mechanized proof system. data NatDed n = Premise (Prop n) | AndI (NatDed n) (NatDed n) | AndE1 (NatDed n) | AndE2 (NatDed n) | Neg2I (NatDed n) | Neg2E (NatDed n) | ImplyI (Prop n) (NatDed n) | ImplyE (NatDed n) (NatDed n) | OrI1 (NatDed n) (Prop n) | OrI2 (Prop n) (NatDed n) | OrE (NatDed n) (NatDed n) (NatDed n) | AbsurdE (NatDed n) (Prop n) | AbsurdI (NatDed n) (NatDed n) | NegI (Prop n) (NatDed n) Using NatDed

• Building a term of type NatDed is a tree-like structure • This tree might be a proof tree. If it maintains the invariant. • A computer program can “check” if that is the case. Constructing proof trees

• Constructing proof trees is a lot like programming. • You are given some premises. These are input to the checker. • You must build a NatDed data structure that relies only on the given premises. • Building this tree is a lot like programming. You must build it out if the constructors of NatDed in such a way that the checker will succeed. Representing the Premises as Data

• data Sequent n = Seq [Prop n] (NatDed n)

Difficulties

• One must think to build a proof tree that will check. • What pieces do you have? – What do they prove? • What other pieces can you make? • How can you put them together. • Sometimes working bottom up helps. • Mechanized help is useful. Strategy

• Construct a term. • Name it. • Let the system check and print it. • Does it prove what you expect? • Did the check complain? • Make some more terms • Put them together. Gentzen style Proofs

• In a Gentzen style proof, we build a tree of hypothetical judgments, instead of a tree of true statements. • Here the set of assumptions (hypotheses, premises) is an explicit part of the proof.

• a |- a ∧ T Gentzen approach

• Here we manipulate both the term to the right of the turnstile ( |- ) and the premises to the left of the turnstile. • This approach is called the sequent calculus The sequent calculus

• The rules are broken into 4 cases. • Some of the cases (the last 2) are broken into left and right variants • The cases – Axiom – Cut – Logical rules – Structural rules Axiom and Cut Logical Rules Structural Rules Intuition

• Logical rules introduce new formula either on the left or the right. They maintain a logical invariant just like the Natural Deduction rules. – What is the invariant? • Structural rules manipulate the formula regardless of the shape or connective that the formula have. Intuiton 2

• Think of the rules as instructions for constructing a proof. • Some of the instructions are ambiguous. There may be many ways to follow them • Next time we will study automated methods for finding a proof.